Slashdot Mirror

Am I the only one here who's goign to ignore everything Intel releases until they start releasing dual-core Dothan-based P4s? Here's Tom's Hardware's take on the new LGA 775 architecture, along with copious comparitive CPU/platform benchmarks. Anandtech has their own entry here. In both cases, the combination of new architectures, cpus, features, etc don't add up to much of an advancement in performance. What you get are a lot of features of questionable value and/or features that have been touted by platforms such as nForce/nForce2 for some time now. Please wake me when the Dothan P4s start coming out.

Let it first be stated that I did not intend to reply to your post in its entirety, as you seem to have assumed. I merely felt the need to correct one statement which I found to be blatantly false. Now, to correct another:

AMD may say what they want, but CPU speed IS the main factor in performance. Because, in the AMD formula above, the "work per clock cycle" is the same for each manufacturer.

That's what Intel marketing would have you believe. But why then does an Athlon FX-53 at 2.4GHz perform better than an Intel Pentium 4 at 3.6GHz, as benchmarked by Anandtech here? According to you, the 1.2GHz difference in clock speed between those CPU's should mean decisive victory for Intel. It obviously does not. Hence, the "work per clock cycle" factor is not equal for different manufacturers. QED.

Regarding your rant about text processors not needing the CPU 99% of the time - that's probably true, but what about what goes on in background processing? I might be compiling Gentoo (or something else), running Seti@Home, virus scanning my hard drive or any number of other background tasks. It adds up. Now, I very rarely tax my CPU in this manner. I doubt there are many people who do. But when it happens, it's nice to have performance to spare, which is why I bought a powerful PC to begin with. As for the people who only use their computer for one thing at a time, they don't need 2+ GHz CPU's in the first place, so they should just go for whichever CPU is cheaper. And as you may know, AMD wins that fight, too.

The ECC logic is broken on the current stepping of the Alderwood chipset.

The feature side of the equation is a lot easier to handle, as Intel has lavished all the features a techie could dream about on the new chipsets. High-Definition audio, Matrix RAID, a new bus with a bright future, and an 8GB per second bidirectional graphics slot are a few of those features that come to mind.

RAID 1 (mirroring) cards are cheap (and sometimes built into the motherboard), but for every gigabyte you want for storage, you need to be two gigabytes worth of hard drives.

RAID 5 cards cost more money (sometimes a thousand dollars or more), but if you set up a 5 drive system, only one of the drives is "wasted" storing redundancy data (of course, you need to buy five hard drives). RAID 5 on three drives is possible, bu the "wasted space" ratio goes from 20% to 33%, as a whole drive's worth of capacity is used to store parity info.

There are other RAID setups that use combinations of striping, mirroring, etc in an attempt to overcome performance bottlenecks.

Another interesting setup is NetCell's SyncRaid/Raid XL - Tom's Hardware had an article on it a while ago, but actually getting it is tough.

If you are planning on building a new system, Anandtech had an interesting article on RAID on the motherboard.

After a few years of experience with Promise RAID 0+1, Promise RAID 5, 3Ware RAID 5 and SCSI RAID 5, and recently 3Ware SATA RAID 5, I would say that the cheaper solutions often provide a false sense of security, especially if using IDE drives. We have a machine that has a Promise RAID-5 IDE setup that on reboot, seems to require a few restarts to get up and running, and when we lost a drive recently, it took quite a while for the array to rebuild (though this might not be an issue at home, where it's faster to rebuild an array than rebuild your whole computer on a fresh drive). I had a Promise RAID 0+1 card in a computer a few years ago that would corrupt any large file that I moved between hard drives on the computer.

If your data is important, get a good card from a trusted manufacturer (3ware is pretty good, and they have open source Linux drivers), and go SATA.

The dual-core chips are, at least initially, going to be targetted at the high-end (and high-$$$) market, ie the server and workstation market where the Opteron currently competes. At least according to this roadmap though, AMD will also be bringing out a dual-core Athlon64 FX chip. This chip could very well be in a socket 939 format (presumably all Athlon64 FX chips going forward will use that platform).

linky linky!

Be careful with HT. These fake CPU's can actually drag down overall throughput due to the fact that they can't do everything a "real" CPU can. My theory is that I/O is one of them. Caution databases servers.

Nope. It's not a "fake" CPU, it's the underutilized part of the same CPU. Since the ultimate limit to a database server is the speed of it's hard drive (and before that and to a lesser extent, the speed (actually *latency*) of it's RAM), hyperthreading can be very beneficial because thread B can execute while thread A is stalled waiting for RAM or disk. Normally, thread A is wasting it's timeslice waiting for I/O *and blocking other threads* but now while it waits, the CPU resources are available to thread B.

See This review and compare the HT vs. non HT scores on the same CPU.

If you're doing x86 stuff, you want an AMD processor.

unless you absolutely NEED to play UT2k4 at 1600x1200 with 4xAA and 8xAF

Pfff. I can do that with a 2-year-old Radeon 9700 PRO on an Athlon 2200. UT2k4 isn't all that taxing.

Yea, but not at 45 FPS. The 9700 Pro on an Athlon 64 3400+ can only manage 16.5 FPS at that resolution & quality.

In all fairness, it's not, in may cases. It is (sometimes) if you're running 1600x1200 with AA turned way up, but at lower resolutions the difference isn't usually at all significant, the ATI 9800 and the NVidia can even beat it at 1200x1024 resolutions. The Nvidia GeForce 6800 Ultra even significantly beats the X800XT in some benchmarks and does it using 32 bit math while the X800XT is using only 24 bit math and older technology in other areas (Shader model 2 vs. Shader model 3, for example).

There's also the question of there being any real gain in having say 130 frams per second for a game as opposed to, say, 85 or 90. Particularly when you monitor isn't likely to be displaying those extra frames (or you eyes seeing them). It would seem to make a lot more sense to just compute and display to to 90 frames a second, and use any remaining CPU power to be sure other parts of the game are as responsive as posiable (and yes, the game does eat a lot of CPU to computer 130 frames a second, even though a lot of the graphics are done right on the video card).

But I'm not really wanting to argue the merits of one card against another, I was just pointing out how the gamers think nothing of dropping $500 here and another $500 in a couple of months when the next "hot" card is available. At this rate they almost certainly will have bought something at least as expensive to replace their new X800XT's before they even have Half Life 2 and Domm 3 in hand, and will be pitching those new cards when they need PCI Expresse video cards for the new motherboards that come out in the fall.

To me it would seem to not make a lot of sense to buy two or three different "high end" AGP cards this year when AGP is about to be replaced by PCI Express in the high end systems this year. But clearly the gamer community is still willing to shell out the cash for anything hot.

And yes, you DID say that Hz was close enough to FPS to compare them

Close enough to compare them in terms of what the eye can see. The statement needs to be taken in the context of my comment (what the eye can and cannot see at what frame rate or frequency), and not as a general statement saying that fps == Hz.

I guess we just have to agree to disagree because I can point to just as many sources that say your theory is bunk as I'm sure you can to my point.

I'll agree to that, but not quite yet. :) I still find it hard to believe that you can run something like FarCry at 1600x1200 4xFSAA 8xAF at a reasonable frame rate on anything but the very latest video cards (reference). Yes, I know that there's absolutely no reason why you would care to run at that high of a resolution, with those FSAA or AF settings, but this ties right back into my claim that "higher fps rates at lower resolutions/detail settings == at least playable fps rates at higher resolutions/detail settings". 15 frames per second at those settings on a Radeon 9700 is unplayable, compared to a somewhat-acceptable 33 fps for a GeForce 6800. At the lower 1280x1024 setting with the same FSAA and AF settings, the 6800 gets a very playable 44 fps while the 9700 is still unplayable at 23. Turn down the resolution or FSAA or AF and I'm sure the 9700 can still hold its own, as can a GeForce 3 (though much worse than the 9700), but that's not the point.

what about the people building thier own computer and selected a smaller processor because it wass afordable and then plan on buying a better one when availible.

Unfortunately, it looks like the Athlon64 platform will not allow us to install "low end" processors into motherboards made for "high end" processors. I read on Anandtech's review that AMD's value line of processors (including an upcoming 32-bit only version of Athlon64) will all be Socket 754 processors while future Athlon64's will eventually all be Socket 939.

What I'm pretty sure this means is that I will not be able to install AMD's Athlon64-based value processor into a Socket 939 motherboard, then install a real Athlon64 4000+ a year or two later when prices are way lower. Even though AMD's "value" processors will be based on the same Athlon64 core, I'm pretty sure socket converters (e.g. slockets, socket370 to fc-pga converters) will not work because of the different on-chip memory controllers (single-channel vs. dual-channel).

So there goes my plan on building a Socket 939 system using AMD's "value" processor, then upgrading to Athlon64 later. Now I'm considering building an Intel Socket 775 Grantsdale-based system with DDR2, PCI-Express, BTX form factor, and a (I'm not kidding) Prescott-based Celeron 325 processor (2.53GHz, 533MHz bus, 256KB L2 cache). By the time I need more processing power, a 4GHz+ Pentium4 (1066MHz bus, x86-64) will be reasonably priced.

Athlon64 64-bit testing under Linux

As introduced with our Athlon 64 article, we are looking at compile times for the Quake III Arena source code. In running this test, we compiled the source for both Q3A and Q3TA in both debug and release modes using the Batch Build feature of Visual C++ 6.0. The build was run three consecutive times for each processor (in order to try to reduce file read time impact), and we used the third compile time.

http://www.anandtech.com/cpu/showdoc.html?i=2065&p =12

It doesn't specify what compiler or platform was used, but at the bare minimum it gives a little glimpse of what you might be able to achieve. Now all you have to do is apply that to a price/performance graph to determine what and how many you want to buy.

-Benjamin Meyer

I've been building computers for quite some time now, and am very glad AMD has prospered with their Athlon 64 line. And, as I've seen with nearly any computer part, this is just the next (if not final) phase of maturity that Athlon 64 will go through. The technology is getting more powerfull, as well as less expensive. And, since this socket will most likely be put on chipsets with PCI-Express and DDR2 support, I think this is a very good time to upgrade. And don't fret- some motherboard manufactures are starting to add backwards-compatibility for AGP and DDR, while adding the new features, such as PCI Express (16x and 1x), and DDR2. I know I'm saving what little cash I have for these products. It's time to upgrade.

Review on Anandtech! (I like them for their print view:)

...I don't think picoBTX is intended for SFF machines, according to an earlier Anandtech article picoBTX is still 8" x 10.5", awfully big for an SFF machine.

Note that the Shuttle XPC requires a proprietary heat-pipe cooling system and a special small, quiet power supply to keep the system cool and quiet. On the other hand, improved cooling and noise reduction are built into the BTX spec. Theoretically, any computer maker will be able to easily build a tiny, cool, and quiet SFF picoBTX machine based on Intel's 6.9-liter reference system.

...I don't think picoBTX is intended for SFF machines, according to an earlier Anandtech article picoBTX is still 8" x 10.5", awfully big for an SFF machine.

Note that the Shuttle XPC requires a proprietary heat-pipe cooling system and a special small, quiet power supply to keep the system cool and quiet. On the other hand, improved cooling and noise reduction are built into the BTX spec. Theoretically, any computer maker will be able to easily build a tiny, cool, and quiet SFF picoBTX machine based on Intel's 6.9-liter reference system.

I believe the boards are all "pico"BTX meaning they're built for Small FormFactor (SFF) PC's. Such as the one's Shuttle Makes

These are picoBTX boards (one expansion slot each), but I don't think picoBTX is intended for SFF machines, according to an earlier Anandtech article picoBTX is still 8" x 10.5", awfully big for an SFF machine.

Shuttle will probably continue using their own custom motherboards for their designs.

Hence the lack of expansion slots.

The microBTX and BTX sizes will have more expansion slots (up to 4 and 7 slots respectively).

OK, what's the deal with LGA-755? Pins on the motherboard? How is it supposed to work? And why??? (the ever changing sockets is one of the things that pisses me off about intel).

PCI-Express 1x. What's the speed? (not much from the looks of it) and what's the advantage over plain old PCI? I'm assuming we will see boards with more than one 16x slot at some point, which would be useful for RAID controllers, gigabit ethernet, and other high bandwidth stuff. But what's the point of 1x slots? Plain PCI works just fine.

BTX. Again, why? And who the fuck came up with this stupid name?

What are the brown slots on these SiS boards?

Am I missing something? They've replaced the standard ATA-IDE connectors with Serial ATA connectors, gotten rid of all of the PCI slots, but for some reason kept the FDD drive connector and the parallel port? Most newer motherboards support booting from USB flash device. As for the parallel port, there aren't many devices being sold today that use them and there are parallel-USB adapters available for those who want to use their old printers.

I know this is a pico-ATX board so it's understandable that the PCI slots were removed for space-saving purposes, but if a pico-ATX enclosure can't fit an extra PCI card, why leave room for a floppy drive?

For more info on BTX have a look at this AnandTech article, or check out FormFactors.org if you want to look at the actual specifications.

They were only showing a picoBTX board. Go read the Anandtech BTX article to see the different BTX sizes proposed. The picoBTX form factor looks about like the small form factor motherboards in Shuttle XPCs. The standard BTX board has the same number of slots as you are used to in normal systems.

• PCI-X: 64bits PCI up to 133mhz
• PCI Express: Serial bus, a replacement for AGP (mostly), and for graphics cards (at the moment). What I accidentally called PCI-X (PCI eXpress)in the article. From other articles I've read, it looks like it'll be used for all cards eventually. Although I still think this acronym makes more sense than PCI-E, I guess it'll confuse some people who have actually heard of PCI-X before. I have this wierd feeling that the industry is gonna call PCI Express PCI-X now (it sounds way cooler).

• There can be different sizes of slots. All the graphics cards use x16 size slots, while it looks like most other types of cards will use x1 size slots.
• It will eventually become ubiquitous like PCI
• More information in this article.

This pic shows the inside of the NV45. Look at the paths on the circuit. Instead of going straight from one chip to another they form different loops, turn around etc. Are they trying to make them longer, or equal distance or introduce picosecond delays or what?

Nice to see both PCI Express x16 and AGP 8X slots on board at least one of them. I'm looking to squeeze a bit more life out of my AGP based ti4200 before updating to one of the newer video cards in a year or so...

I think this picture shows you exactly about the size of the new Nforce 3 boards with many PCI slots. I'm really hopeful that these new boards can perform as well as the last ones.

AnandTech has a halfway decent article on the spec. Good summary of what it is and who it's aimed at. As I understand it will eventually be mainstream, where ATX will be phased out just like AT was.

If you don't feel like clicking through a hundred pages, use the "print" link instead.

The system is basically an accelerometer which monitor the movements of the laptop, and spins down the HD when there is a risk of impact. I would like to write a Linux driver for it, but I refuse to reverse engineer the windows driver. More info here

AnandTech usually does them in their processor reviews, lemme dig one up.
Here's one, for example.
(Of note, the Athlon FX-51 and -53 are identical to Opteron 148 and 150 processors, respectively. The Athlon 64s are similar as well, difference is they use a different socket, have only single-channel memory controllers, and use unbuffered/unregistered memory.)

Basically, the Hammers are godlike at compilation.
The lowest-rated (at the time; a 2800+ has since been released) A64 3000+ beats the fastest P4 3.4GHz Extreme Edition.

Here is a review of the new X800 card. It seems to have the PS2.0 like you say, but with extensions. Other than that it's comparable, but faster (raw pixel pushing and such), than the new nVIdia card. Now, I prefer ATI cards for their raw speed, but I acknowledge that nVidia's card may possibly be faster or prettier in the new engines.

Power dissipation under load is similar between AMD and Intel chips.Anandtech shows the idle temp of the athlons a little higher, and vice versa for the load temps. There's around a 10% difference using the same heatsink.

Why the assumption that most people don't use their cpus? Mine is at 100% a lot of the time either running tests for development work or gaming.

I've never actually owned a PC, so it might be a good learning experience to put one together... hrm.

Don't know if drivers was a typo, but what you said may be more true than you realize - the switch to writing to DL (dual layer) DVDs is looking like it might be accomplished with a simple firmware upgrade, at least for some drives. Check this out for a little more info (basically, a new DL writer appears to use exactly the same controller as a single layer writer, so it looks like the difference is largely a firmware thing).

A recent article showed that the DL write speeds at 2.4x. So you can spend 45mins burning 1 DL DVD, or 2x 15min burning two DVDs at 8x.

Actually, the GeForce 6800 includes the hardware to do just that. I'm surprised no one else has mentioned it by now, as I thought it was one of the cooler features of the new NV40 chipset.

That's a big assumption. AMD has always tried to keep a per-die cost lead on Intel. They do this by keeping their die-size down so that they achieve higher yields. That's why most of AMD's future low-end Athlon 64s have 1/2 as much L2 cache as the Pentium-M or P4 Prescott.
Actually Intel's adoption of 90nm & especially 300mm wafers make most analysts believe they have cheaper production costs. AMD before had smaller die size on athlons to compensate for 200mm wafers, but the athlon64 is 192mm^2, while pentium-m is 82mm^2 and the next gen pentium-m will be 84mm^2.

The PIII was not a good match for RDRAM. RDRAM had a faster transfer rate but more latency than SDRAM. With the chipsets of the day (i815, Apollo Pro 133A, i440BX overclocked to 133) PC133 was faster in most benchmarks. See for yourself, Anandtech still has their article online. By the time the P4 rolled around, it was a better match for RDRAM. The i850 boards were significantly faster than the SDRAM and DDR boards of the day, but by then most people wanted nothing more to do with Rambus.

AnandTech review on a full page and without much ads

stop clicking the "next page" links every paragraph and try this out! anandtech.com review [anandtech.com]

Really? Checked out the price of a P4 Extreme Edition CPU lately? Check it out:
Intel Pentium 4 3.2GHz 800FSB 512KB: $279.00
Intel Pentium 4 3.2GHz 800FSB Extreme: $910.00

Intel Pentium 4 3.4GHz 800FSB 512KB: $412.00
Intel Pentium 4 3.4GHz 800FSB Extreme: $1,139.00

A grand for a CPU... man, I thought those days were long over...

Need more horsepower... the Opteron 4-way boxes (HP 4-way), crush the Intel Xeon's (as do the two ways) in most web and DB benchmarks. Oh yeah, they are usually priced comparably or cheaper than the Intels as well.

Need more horsepower... the Opteron 4-way boxes (HP 4-way), crush the Intel Xeon's (as do the two ways) in most web and DB benchmarks. Oh yeah, they are usually priced comparably or cheaper than the Intels as well.

Heat comparison

Check out that "Load Temperature" chart. What's that? Intel's at the top and the Athlon's are at the bottom? Even at idle temps, the Athlon64 comes in under or even with the P4s.

To clarify on a few points of the parent:

The nVidia 6800 Ultra requires two dedicated molex power connectors, and it also requires a 480W power supply. More details. Now that's a lot of power.

Also, the cooling setup on the 6800 Ultra takes up a slot of its own, which means you lose a PCI port as well, although now that most of the features PCI had (such as sound and NICs) are integrated in the motherboards, it's not too big of a deal.

Lastly to note, nVidia is releasing a lower-powered 6800GT which is approximately equivalent to the X800 Pro card, and they just recently announced a 6850 Ultra which is basically an OEM-overclocked 6800 Ultra. That thing will probably take up 5 slots, have a built-in A/C unit, and have its own cold fusion reactor as well.

Bannerless/click-through-10-page-less:

HotHardware

AnandTech

stolen from Anandtech

HardOCP
Ascully
DriverHeaven
TrustedReviews
K-Hardware
Hardware Analysis
Hexus
The Tech Report
Beyond3D
Neoseeker
ExtremeTech
Gamers Depot
Lost Circuits
Firing Squad
Tom's Hardware
Bjorn3D
Hot Hardware

Your comment has too few characters per line (currently 10.9). Your comment has too few characters per line (currently 12.3). Your comment has too few characters per line (currently 14.9). Your comment has too few characters per line (currently 17.4).